The Igneum whitepaper

Earn, spend, bond

IGN is earned by GPU work, spent on transactions and EVM applications, and locked by keys selected to sign blocks. On the development chain today, it moves through the wallet and Pay; no mainnet coin or market exists yet.

• GPUs earn IGN.
• Users spend IGN for gas and payments.
• Applications use IGN as the native asset.
• Signing keys bond IGN for admission to the draw.
• Base fees remove IGN; the fund finances disclosed public work.

What IGN is for, today

The four numbers, the record of 13 October 2026

The project set itself four numbers before the measurements.

On the record of 13 October 2026 three of the four stand, two measured and one a model; on the registered programme the energy line is met at 7.9 and 7.2 microjoules a hash, and the chip question is answered by the chain's pay, not by one lane: a chip built for one lane earns 1.2x to 1.4x a card at the weights of record (0.2 programme, 0.4 bandwidth, 0.4 raster, on the research chain from block 6523), the stacked chip 1.6x disclosed, and 1.3x to 1.4x at thirds, the history; the sealer's quarter is folded into the pot from block 6666 on the research chain (15 October 2026, 22:38 UK), a sealer paid only on its own standing; the public test network takes it by its own registry entry when that lands. The goal of record: a simple, efficient GPU workload whose complete specialist implementation saves too little to justify building and buying it. The full record of each is on its deep page.

Abstract

IgneumHash is implemented and exact on tested NVIDIA, AMD, Intel and Apple hardware; consensus and GPU-worker succession round trips pass on the tested pairs.

Igneum is a chain not yet at genesis whose current mining relation has exact implementations on tested NVIDIA, AMD, Intel and Apple GPUs. It implements an EVM-compatible execution environment; the supported compatibility surface remains subject to the published application test suite. Its proof of work, IgneumHash, is built so that the dataset item a miner fetches is itself the programme the lane runs next: each fetched item also defines programmable lane work, so a specialist must provide both the dependent memory path and a programmable lane array. The 11 October model read 2.59x centrally and 3.25x at the governing upper; it was conceded on 12 October 2026, and a specialist chip's edge on the registered relation is under measurement on version 14.1; the figure is published when it reads. The late-Ethereum precedent sits below the corrected central and does not guarantee Igneum's outcome. The figures are a model until an externally placed design exists, and the page that carries them says so.

Two more rules hold the chain. Work authority, live from block one, makes the chain hybrid: proof of work determines share-lane rewards, bonded draw units select signers where units are occupied, and the open-seat proof-of-work rule supplies liveness while units remain empty. An unbonded share worker cannot alter the signed template without invalidating its work; the selected bonded key supplies the block seal. Hashing earns; the selected key signs. The rule's three rows (outsourcing, buying the draw, pool control) closed on the record of 11 October 2026, each with what it leaves open named; final launch qualification awaits the registered clock, fork, liveness and adversarial rows. Succession replaces the relation on a schedule fixed in advance: the proposed competition uses predeclared admission and scoring rules, and autonomous selection is not claimed until its contract and no-candidate path pass. The machinery has been exercised warm on a live two-node chain and on a GPU worker with no rejected or invalid work.

Every figure in this paper carries the grade the record gives it. Measured means a stored row on a named instrument. Model means arithmetic on named inputs. Where a row is not final, the paper says what it awaits. Nothing here is a launch. No Igneum mainnet coin currently exists.

Where it stands

15 October 2026. The public test network runs on a new chain from its genesis since 15 October 2026 at 17:21 UK: chain id 7322, the three lanes paid from genesis; the public test network moves to 12-second blocks tonight with 0.10.40 (the research chain's measured hour on 16 October, 16:05 to 17:05 UK: 292 blocks at 12.3 s mean, none replaced). The old public test network (chain id 7320), retired 16 October 2026 at block 15,371, sealed from 12 October 2026 at 23:41 UK and carries the record: eight single-fault releases without a reset, one canonical history replayed by nine independent readers, every accepted share paid exact to the wei. The research chain pays three lanes, the bandwidth lane, the raster lane and the programme lane, at the weights of record, 0.2 programme / 0.4 bandwidth / 0.4 raster, from block 6523 (equal thirds from 5063 to 6522) by registry transaction (each landing on the registry page); the pool pays only the lanes the chain pays, and the breadth rule's opening payment is replayed to the wei. On the registered programme the energy line is met by our own hand: 7.9 microjoules a hash on a 5090 (75.8 MH/s at its 600 W cap) and 7.2 on a 4090 (35.6 MH/s at 257 W) by the registered form's compiled kernel, exact on NVIDIA, AMD, Intel and Apple on its registered sets (10.8 and 10.0 by the like-for-like kernel at the stock caps, the basis of the old-card ratios), verified on the minimum node in 82 ms a 256-entry list at the median, the slowest percentile 143 ms disclosed, a pool key paid under the breadth rule to the wei. A chip built for one lane earns what a card earns. Against the denominator of record (the 5090 FE at its 450 W cap) the model's reconfiguring fabric reads 4.2x on energy on the programme lane, the board chip 2.41x on the bandwidth lane, the raster chip 3.2x on its drawing line and the stacked machine 10.2x, disclosed; the earlier set on the pod's 5090 at its 600 W cap and the 4090 at 257 W (the fabric 4.6x and 4.2x, the programmable chip 2.9x and 2.6x, the stacked 10.2x and 9.3x) is history, dated 15 October; KawPoW 7.3 and 5.5 microjoules a hash by the same hand. The chain pays three kinds of physics equally and a key is paid across them, so a chip built for one lane earns 1.2x to 1.4x a card at the weights of record (0.2 programme, 0.4 bandwidth, 0.4 raster, on the research chain from block 6523), the stacked chip 1.6x disclosed, and 1.3x to 1.4x at thirds, the history (the sealer's quarter is folded into the pot from block 6666 on the research chain (15 October 2026, 22:38 UK), a sealer paid only on its own standing; the public test network takes it by its own registry entry when that lands), against the model's line of about 1.3x: a chip that at most breaks even and is not built. The chip left is a box of three designs, 3.4x on a programmable basis and 5.4x on a fabric basis, an estimated 19.8 million pounds a design, disclosed; and if one comes, the registry moves the lane's form by one transaction. Measured, not argued; the chip-case figures are a model until a design exists. On the exact rows a 2060 earns about 0.5 of a 4090 on the bandwidth lane at stock settings (the rented 2060 at its 130 W cap 0.51, e929c30c9; the owned 2060's own stock row 0.465 by NVML and 0.444 by the wall, 785078dc5, its samples 39d3d42e0) and 0.24 on the programme lane at its stock cap (26ef222ef), and a 3060 0.69 on the bandwidth lane at 131 to 133 W (row 26 of the version 15 table, 4e48be3b). On the raster lane, like for like by the card's counter, the 2060 reads 0.99 of the 4090 an attempt by totals and 0.84 by increments, and the 3060 1.10 by totals and 1.40 by increments on the same text (72e4920ef, 43aa36aab); the 9060 XT's 50 mJ an attempt (f27881574) and the M6's 26 mJ (a port figure, 066544bd0) are plug figures with no counter row on the 4090 to read against, served as energies with their grade and no ratio. The candidate successor form, entry 2, reads 0.27 on its exact rows. Ember, the miner, mines, tunes and pays on Windows, macOS and Linux as a signed release that updates over the air, with a 1 percent fee you can switch off beside the pool's 1 percent.

What is still open, served in plain words: the raster lane on its drawing line until version 19 is mined, its rows port rows until then; the entry 2 successor form's rows on the old cards; and the chip-case figures, a model until a design exists. The record lives in the repository's record and in the status feed at build.igneum.network/feed/status.json; each open item is published as it moves.

The problem

Graphics-card mining lost its largest home when Ethereum left proof of work in 2022.

Since then the chains that kept mining on cards have gone two ways: a chip arrived and took the work, or the chain stayed small. Two things make a chip arrive. A fixed relation, so that a chip built once stays right; and a relation whose work is a plain memory walk, so that a chip is a controller and the memory of least cost that holds the dataset. The record of 10 October 2026 read such a chip, for the rejected predecessor relation (a plain memory walk, not IgneumHash), at about 3.9 times a current card for an external-memory board and about 7.8 times for a hybrid-bonded stacked-memory device; the second failed the launch line, and the predecessor was retired.

A chain can answer the chip in two places: in the relation, by making the work something a memory chip cannot do; and in the rules, by bounding what a hasher controls regardless of how much it hashes. Igneum tests both approaches: IgneumHash addresses the specialist's architecture, while work authority has its three rows closed and its registered rows ahead. Each is measured against a standard written before the readings.

Igneum at a glance

Mining: the item is the programme

A miner holds a dataset (2 GiB at the gate cells, the launch default) and walks it: each step fetches an item that depends on the one before.

No prediction shortcut has passed the registered attack tests; the million-nonce distribution and grinding rows remain part of qualification. Rate is set jointly by the dependent memory walk and the programme executed from each fetched item. The fetched item is the programme the lane runs: a block of single-precision operations with fixed rounding on every opcode, a cross-lane class in which lanes exchange values through the register crossbar, a 16-register lane window, and fixed quotas on the opcode classes. Fixed quotas remove opcode-count variation; the registered grinding tests determine whether any cheap-item selection remains. There is no generator: the earliest measured rows showed a generated programme cost fourteen times the walk. The verifier recomputes one item's programme and reads inside 2 ms on the node's path.

Each lane carries a registered share of the pooled pay. On the research chain the three lanes, bandwidth, raster and the programme, are paid at the weights of record, 0.2 programme / 0.4 bandwidth / 0.4 raster, from block 6523 (equal thirds from 5063 to 6522) by registry transaction, the earlier lanes registered at no share; a key is paid on its weakest lane under the minimum rule, live on the research chain from block 5958. A key that mines every paid lane is paid by the chain directly, to the wei; a pool's members mine under the pool's key, the chain pays that key each window, and the pool pays each member its due from that payment, the table public and checkable against the chain. A share is a registration: it moves by a signed transaction with its notice, and the registry key rotates by the same path; the six landings of today are on the registry page. The public test network runs on a new chain from its genesis since 15 October 2026 at 17:21 UK, with the same three lanes from genesis.

The launch document is admitted. Its form is version 8 of the relation with the bond-keyed rounds: in two of the walk's 64 rounds (four in the alternate setting) a share computed from the sealer's bonded secret enters the fetched item before the lane's work, so a block's seal cannot be produced without participation by that secret, while the shares anyone submits to a pool stay unkeyed. On 11 October 2026 the admission rule's two-implementation test was run on both settings: the reference implementation and an independent second implementation, written in another language from the specification's text, agree on every class of both frozen vector sets, including the malformed, trace and search classes, in both directions of checking; the verifier's time on the node's path is 0.785 ms at the median, under the 2 ms line. The relation reads ADMITTED on both tags.

Every implementation is exact against one scalar reference and one frozen vector set per gate document, for every registered gate document. The rule for every vendor's kernel is one shared source text: no runtime-indexed private array and no conditional expression inside a hash routine, with per-machine exactness sets on every supported compiler class. The three-vendor reading, the rows behind it and their boundaries are on IgneumHash.

The cohort of record, with its grades

The cohort record of 12 October 2026, 20 rows; every ratio reads against RTX 5090, rented pod, 500 W cap: 40.6 MH/s and 12.20 uJ per accepted hash.

• NVIDIA, RTX 5090, rented pod, 500 W cap: 40.6 MH/s and 12.20 uJ per hash, measured (board telemetry), 11 October 2026.
• NVIDIA, RTX 5090 Founders Edition (bench PC A), owned, 450 W cap: 37.0 MH/s and 12.06 uJ per hash, measured (board telemetry), 11 October 2026.
• NVIDIA, RTX 5090 Founders Edition (bench PC B), owned, 575 W cap, free clock: 13.44 uJ per hash, measured (board telemetry), 11 October 2026.
• NVIDIA, RTX 4090 (bench PC B, enclosure 2R), owned, 480 W cap, uncapped: 51.4 MH/s and 8.38 uJ per hash, measured (wall differential), 12 October 2026.
• NVIDIA, RTX 4090 (bench PC B, enclosure 2R), the 380 W knee, owned, 380 W cap: 46.3 MH/s and 7.05 uJ per hash, measured (wall differential), 12 October 2026.
• NVIDIA, RTX 4090, rented pod, 400 W cap: 46.64 MH/s and 8.531 uJ per hash, screening (board telemetry), 11 October 2026.
• NVIDIA, RTX 5080 (bench PC A, enclosure 1L), owned, 360 W cap: 22.9 MH/s and 13.54 uJ per hash, measured (wall differential), 11 October 2026.
• NVIDIA, CMP 170HX (the rig, as shipped), owned, 250 W cap: 17.1 MH/s and 11.30 uJ per hash, measured (wall differential), 12 October 2026.
• NVIDIA, RTX 2060 6 GB (the rig, the owned floor-class card), owned, 172 W cap: 7.9 MH/s and 17.60 uJ per hash, measured (wall differential), 12 October 2026.
• AMD, RX 9070 XT (bench PC A, enclosure 2R), the card of record, owned: 15.5 MH/s and 13.34 uJ per hash, measured (wall differential), 11 October 2026.
• AMD, RX 9060 XT, the corroboration, owned: 14.24 uJ per hash, measured (board telemetry), 11 October 2026.
• AMD, Instinct MI300X, rented pod, 750 W cap: 47.93 MH/s and 11.73 uJ per hash, screening (board telemetry), 11 October 2026.
• Apple, M6 mini (16 GB), owned: 1.97 MH/s, estimated, 11 October 2026.
• Apple, M5 Max (64 GB, the 14-inch laptop), owned, 100 W cap, High Power: 5.4 MH/s, estimated, 11 October 2026.
• Intel, Arc B580 (bench PC B, enclosure 2R), owned: no rate or energy row yet, pending, 11 October 2026.
• NVIDIA, RTX 3060 12 GB, owned, 170 W cap: 8.5 MH/s and 19.90 uJ per hash, measured (board telemetry), 12 October 2026.
• AMD, RX 580 (the rig), owned, the generic OpenCL path, untuned: no rate or energy row yet, pending, 12 October 2026.
• AMD, RX 7600 (the rig), owned, the generic OpenCL path (Mesa rusticl on radeonsi), untuned: 1.01 MH/s, measured (none), 12 October 2026.
• AMD, RX 5700 XT (the rig), owned, the generic OpenCL path, untuned: no rate or energy row yet, pending, 12 October 2026.
• AMD, RX 6600 (the rig), owned, the generic OpenCL path (Mesa rusticl on radeonsi), untuned: 0.94 MH/s, measured (none), 12 October 2026.

The chip question

Igneum does not claim to keep specialist hardware out.

It states a standard, registered before the measurements, and reports each reading against it with its grade.

Why the model of 11 October predicted a smaller specialist edge, since conceded: a shipped memory stack has no lane array and cannot run the programme. The chip that can is a new GPU-class die with no incumbent, 35 to 150 million USD of non-recurring engineering; its clock from decision to volume, by the adjudication of 12 October 2026, 13 to 15 months central with the band 9 to 22 (a new entrant with a GDDR6 board 20, an incumbent mining-ASIC vendor 13, a stacked-DRAM incumbent 15, an FPGA bitstream 3), estimated. On the captured trace a single lane runs at a fifth of its capacity, because the item fetch dwarfs the programme block; reaching full utilisation needs eight interleaved contexts and a 128-entry register file per lane, dearer than the extreme figure assumed, so the chip's two favourable assumptions cannot hold at once. The model's external placement is still outstanding. The standard's history, including the line it retired and the failure that stays on the record, is on The standard.

The complete standard is not met today. Two guards read inside their lines on a model; the third has its three rows closed, with what each leaves open named, and its launch qualification follows the registered clock, fork, liveness and adversarial rows on the next development chains.

Work authority: hashing earns; the selected key signs

From block one the chain is hybrid. Proof of work determines share-lane rewards. Bonded draw units select signers where units are occupied; the open-seat proof-of-work rule supplies liveness while units remain empty. Anyone with a supported GPU may submit valid shares without posting a bond, and accepted work receives its pro-rata portion of the block payout under the published share target. A qualifying bond receives draw units under the published allocation rule and becomes eligible to sign when one of those units is selected; the bond is per key, and no verified person, company or supplier stands behind a key in the protocol. The selected bonded key must participate in producing a valid key-bound proof-of-work seal. A seal without a valid proof is consensus-invalid; the proof establishes knowledge of a witness linked to the bonded public key and the keyed seal relation, subject to the registered proof system. What the three rows showed, each with what it leaves open: remote sealing over a link is bounded at the S=4 launch setting (below 10 percent of draws won at 95 percent confidence on both a consumer and a datacentre link), while a co-located appliance, a holder-operated sealing service and a builder market remain possible and are disclosed as open economic questions; splitting one bond across any number of keys gains no expected draw authority (the chain prices split identities neutrally; it does not attribute suppliers); the member-node pool cannot replace the block, the seal, the payout list or the reward key, combine members' draws, or withhold and redirect a solution, while custodial pooling, a pool holding members' keys, remains open and disclosed. A bond is locked for the entry (twelve months). An equivocation, two sealed blocks or two endorsements under one key at one height, forfeits the whole bond under a rule built in the node and exercised on the development chains; its consensus rows on the launch-clock chain are a launch gate. The bond's reference level is 16 percent of the coins issued at the year-one boundary; whether that much is bonded is an outcome the chain measures, never a rule it enforces.

What it binds: who may sign at a height; what an unbonded share worker can change in a signed template (the nonce and nothing else; any other change invalidates its work); where the reward goes; which parent a block extends. What it leaves, stated: hash share, custody, and the off-chain sale the chain cannot see. The rule, its frozen parameters, the rows and what they leave open are on Work authority.

Succession: a process fixed in advance

The chain has no fully implemented autonomous successors at launch.

It has a proposed process and timetable for selecting them: a competition under ten locked rules, with the schedule fixed in advance, incompatibility with the incumbent die mandatory, the card cost bounded by the supported-card floor registered for that competition, objective scoring frozen before submissions, a defined no-candidate outcome, and past consensus immutable. The proposed competition uses predeclared admission and scoring rules; autonomous selection is not claimed until its contract and no-candidate path pass. The machinery that moves the chain from one relation to the next has run: seven height-activated transitions on 10 October 2026 with no manual intervention, then the round trip from the launch document to IgneumHash version 8 and back, warm, on two nodes and on a CUDA worker, in both directions with no rejected or invalid work, while paying a 12 percent rate reduction during the 32-block preparation window; the boundary swap itself added about one second.

What succession does not do: reset the clock on a programmable die. That question is unresolved as ruled, and it is why the long-clock guard reads the chip's lead and not any succession. The RTX 2060 is named nowhere as a gate: the owned 2060's row on the rig is the measured floor-class cohort row. The rows and the rules are on Succession.

The chain of record

The public test network: chain id 7322, network id 7394, opened 15 October 2026 at 17:21 UK from its own genesis (0x45828a34b05153fd2a9d0b7c49e84ec2289c38dd2f8c755729321e4f9c2236ea) with the three lanes paid from genesis; its public RPC is rpc.igneum.network, its pool pool.igneum.network:3342, its status feed build.igneum.network/feed/status.json. The old public test network, retired 16 October 2026 at block 15,371 with its legacy name serving reads only: chain id 7320, network id 7392, opened 12 October 2026 at 23:41 UK from its own genesis with the accumulated clock, fork, share-accounting, release and epoch-transition fixes under one reproducible build. It sealed a block every twenty seconds at its launch constants (the sealer's time floor at 12 seconds, its proving allowance at 250 milliseconds), the open seat and the floor-priced unit live from block one, the share lane and the endorsements live from genesis on independently built nodes. Its release sequence is the record's: eight single-fault releases activated by height without a reset, each announced with its binary's sum, the rollback target named beside it; its history replays on nine independent readers; the 100-block outage exercise that preceded it showed one-history continuity, exact credit and ordinary peer-sync recovery. Its legacy name serves reads only; the explorer follows the public test network.

The project's stance, 11 October 2026: the mining result stands on the record at the grades the record gives it; the network, products, economics and ecosystem now have to demonstrate that they deserve it.

Beside the mining side stands a list of fourteen items, each with a bar and a test on the record: consensus and settlement; economic security at the floor; fair share payment; the production miner and pool; node affordability and chain growth; supply, fund and accounting; wallet and custody; EVM and contract compatibility; the public builder platform; grants, maintenance and security response; governance and succession; fair launch, communications and anti-fraud; independent usage; and market access. The launch readiness ledger carries each with its rows and their state; none reads done.

The default mining endpoint for Ember and the wallet is the company's pool at a public address in the member-node form: the member's own node seals its own template with the pool's payout list, so the coordinator can replace nothing; the pool's disclosure document and its share window are published beside it. The pool's fee is 1 percent of what you mine on every session; Ember's own fee is 1 percent, shown on its Earnings page, and you can switch it off. Solo mining against your own node over the published mining RPC stays the earliest option in every list.

Economics

This section is the economics text at its close of 11 October 2026, taken whole; every figure in it is from that close and the record of the day.

Issuance

Igneum is a proposed chain whose mining relation has been implemented and tested on NVIDIA, AMD and Apple GPUs. Coins come into existence only as block rewards: there is no premine, no presale and no allocation to anyone at genesis. Blocks come every 12 seconds (the public test network moves to 12-second blocks tonight with 0.10.40). Every epoch of 6,144 blocks (about 20 hours at 12-second blocks, until the 12-second re-cut, 0.10.43, gates the epoch to 10,240 blocks and the subsidy a block to 12/20 of the 20-second figure, the coins a year unchanged) issues one part in 2,048 of what remains below 100 million IGN, the declining phase's target, paid evenly across the epoch's blocks with the exact remainder in the epoch's last block, so the supply approaches the target smoothly, with no cliffs: the first epoch issues 48,828.125 IGN, about 7.95 IGN a block. Half of the target is issued in the first six years. From about year 25 the reward settles at a fixed 2,048 IGN an epoch, one third of an IGN a block, 525,600 IGN a year, about half a percent a year of the supply by then and falling as a share, and stays there. The alternative, a reward that keeps declining to nothing, was built and tested beside it and was not chosen. The launch protocol contains no vote, price trigger or forecast mechanism that can change the issuance rule. Two independent implementations of the schedule, one in the node and one written from the rule's text, agree to the smallest unit at every epoch boundary to year 60, and a live development chain reads the same figures.

Where each block's reward goes

Of every block reward, 99 percent goes to the people who mined the block and 1 percent to a maintenance fund. Transaction fees have two parts. The base fee is burned, as Ethereum's, from genesis on every Igneum chain (the node's code paths at 27bd7bf88); the tip goes to the block's sealer (under a payout list it is settled with the subsidy by the block's shares split); sponsored-lane fees are a separate path to the miners on exact share, a future design, not code. Every transaction pays a base fee per gas, set block by block by the chain's own rule with its constants fixed from genesis and never falling below 1 gwei; the base fee is credited to no one and leaves the supply. What a transaction pays above the base fee is the tip: the block's producer is the account that sealed the block (a bonded key, or the open seat's producer while units stand empty), or, when the block pays a pool's members by its payout list, those members in proportion to their listed weights, together with the subsidy. A pool's fee is whatever weight its coordinator gives itself on that list, visible on every block the pool seals; there is no other fee path. Anything that reaches a list's address between two blocks paying the same list is swept into the next such block's payment. The maintenance fund takes its share of the block reward only, nothing from fees or penalties.

The maintenance fund

The fund is the network's; no person or company owns it. Its mandate names five heads of spending and no other: client maintenance, security review, infrastructure, independently evaluated grants, and open-source protocol and client development. It may not hold any other asset, lend, buy hash rate, bond for a signing seat, vote, or pay anyone for holding, for a price or for a listing. It is held by a three-signer contract that spends at two signatures, freezes at one, grows to seven signers by published additions with sixty days' notice each and quorums that rise with the size, caps spending at a quarter of its balance in each ninety-day window unless a sixty-day notice was published, records every proposal, approval, recusal, payment and refusal as a public event, and has no owner, no upgrade path and no pause beyond the freeze. The three opening signers, published by initials: J. M. (the company), S. S. (independent), O. W. (independent); their keys follow the publication. Signers connected to the company may never be a majority of the set or of any payment to the company, and a signer with an interest in a payment stands aside from it in public. The fund publishes its signers and their affiliations, every payment with its scope and delivered work within thirty days, quarterly accounts and an independent annual review. The contract is deployed on the development chain; the signer set is to be published.

The company's money

Igneum Labs receives no automatic protocol allocation: it earns by mining under the public rules, through Ember's visible optional 1 percent fee and other declared services, and through published contracts for work approved by the independently governed maintenance fund. Ember, the reference miner, carries a 1 percent software fee that is shown on its earnings page beside the fund's 1 percent and can be switched off; it is software, not the chain, and a miner who switches it off loses nothing. The company's pool charges 1 percent of what you mine on every session, and Ember's own 1 percent is shown on its Earnings page and switchable; the fund's one percent of every block is the chain's rule, not the company's. No team cut means exactly this: the chain itself hands the company nothing automatically, there is no hidden slice, and there is no switch to add one later; the chain's rules contain no developer share and none can be turned on after genesis.

What changed from the earlier plan

Earlier texts described a hard cap and a chain with no fund and no team fee. The permanent tail replaces an ordinary hard cap: 100 million IGN is the declining phase's target, never a maximum, and what survives of the old rule is that no discretionary, vote-triggered or forecast-triggered issuance exists. The old "no fund, no team fee" sentence is retired for the accurate one: 1 percent to a maintenance fund with a public mandate, which is not a company allocation, and the company earns only as stated above. The three decisions behind this, the perpetual tail, the 99/1 split with the fund's mandate, and no automatic allocation to the company with the developer share permanently off, were ratified in writing by the person who took them and read by the outside review as meeting its requirement for a monetary freeze.

Mining without a pool

Solo mining requires no pool and no bond. A supported GPU may submit shares directly to a node, and accepted work receives its pro-rata portion of the block payout under the published share target (160 accepted shares a block, 256 the ceiling; a share whose parent block leaves the chain in a reorganisation expires and is named so per address; every other accepted share is paid in the block or carried forward). Payout frequency varies with card rate, network work and list capacity. No daily payout or income is promised.

The gates

The money design is measured against thirty-two gates taken from the project's own token-value standard, and none of them passes in full today. One, the reproduction of the supply schedule by two independent implementations with a live chain read, passes in its launch-relevant scope. Three more pass on named clauses and wait on the rest: the fair-launch allocation clause (nothing allocated at genesis), the fee reconciliation (every payment on the development chain reconciles to the smallest unit, with the share lane's lossless accounting proven per address and its repeat on the next chain pending), and the bond's tests (Sybil-neutral pricing, the objective forfeit, the unit and the floor read back on chain). The security-budget gate has its model and every cost cell on the record, each a stated bound, and reads as a failure region rather than a pass, because the purchasing power that would turn costs into a verdict is not something a chain without a market can supply. Thirteen gates cannot be run before genesis by construction, because they need a market, users, customers or products this design does not have, and they are labelled so rather than claimed. The remaining gates each wait on a named measurement with a named owner. Nothing in this paper is a claim about price, value, return, adoption or rank.

When a supply figure is shown, five are shown, separately: total issued, burned, bonded, in the fund, and liquid circulating. No page redefines the circulating supply by silently subtracting the bonded or the fund's balances, and the 100 million is the declining phase's target, never a maximum. Igneum makes no promise about price, profit, yield or market ranking.

For miners

One command on Linux and HiveOS, an installer on Windows, a package on Apple silicon.

Ember finds the cards, runs one worker per card from its own install directory with the worker's checksum read against the release manifest, tunes each card to the knee, and mines the company's pool by default or your own node. It shows the rate, the energy per hash the card reports or the wall figure if you entered a meter, what a day costs at your tariff, its own 1 percent fee beside the fund's with the switch to turn it off, and every accepted or rejected share with its reason. Ember 0.2.21 is the current release: 8 packages for The source tree, The mirror on GitHub, Windows, macOS, Linux and HiveOS, every package signed under the release key, served under its release directory with SHA256SUMS and its signature.

Solo mining is the earliest option in every list: your worker against your own node over the published mining RPC, or the pool service in solo mode. Independent pools run the same service. Every rejection carries a word you can read. No rate table by card and no income figure appear before the release; when they do, every figure names its row and its grade. The rest is on Ember.

The wallet

Three processes, one window.

The wallet is designed so that the window host holds nothing secret; the engine holds the public state and talks to the node; the signer holds the vault and the unlocked key in locked memory, is designed to talk only to the engine, and is designed to have no way to reach a website. On Linux the signer runs as its own process, isolated from the page process, with no network reach and nothing released to a page before the person's yes (rows ISO-01 to ISO-03 and the suite's 34 rows pass on the signed 0.0.4's own binaries, read against the chain and the binaries by the independent security seat), and seven participants on stock Linux machines, from the public pages alone and without help, have received, sent, backed up, recovered on a second machine, and bonded and exited with the packaged releases, every step read from the chain. Nothing on screen calls a transaction final: the display says submitted, included, and how much work sits on top, and the word settled appears only against a policy you chose and can read. A website is designed to get nothing until you say yes in the wallet window, and every request is shown in words before it is signed. The payment contracts have no owner, no upgrade path, no pause and no protocol fee. The Igneum wallet is the current release: 3 packages for macOS, Linux and Windows, every package signed under the release key, served under its release directory with SHA256SUMS and its signature. The rest is on The wallet.

Governance

Every choice the project made is in a decision register with a status (proposed, ratified, frozen, or kept open) and its reason; a ratification is a dated line with the exact wording and what was left open, written by nobody but the person who took the decision. The register has no calendar timeframes: phase order and gates only. The launch standard's decisions (the amendment, the route, the entry length, the card floor) are entries in it. After genesis, a change to a frozen parameter is a new ratified row under the selection contract, never an edit; a consensus fault is a fix row with the block hash and both nodes' verdicts, never a supplier-share estimate or a signed word. The register is on Build on Igneum.

The road to launch

There is no date. The launch gates are seven rows, every one on the record before a genesis date is fixed: the three rows of work authority are closed and verified (what they leave open named; launch qualification on the registered rows); two independently built nodes agree on every block from genesis across three lottery epochs; a fresh node syncs across the same; the vectors and binaries are frozen and reproducible on two builders with the parity row under the keyed form, and the relation is admitted by the two-implementation test (met); the keyed form costs at or under 5 percent of the hash's energy on the measured 5090 row (met at the launch setting); settlement by endorsements passes its consensus rows; and the share lane passes its seven rows. Exposure is counted from the earliest actionable public disclosure, a genesis delay reduces the active lifetime day for day, and the retirement height is committed at genesis. What is final already: the warm round trips of succession. What is a model or still open: the small edge and the long clock (a model pending an external placed design), Apple on its wall figure for the admitted relation, the production worker on longer repeats. AMD family qualification passed on the RX 9070 XT at telemetry grade; wall confirmation remains. What still awaits: the registered clock, fork, liveness and adversarial rows of work authority on two earlier development chains, retired 16 October 2026.

What you can verify

The fleet feed, the attribution diagnostic and the health probe are served as JSON at public URLs; each answered without credentials when checked on 11 October 2026 at 10:36 UK, and a row stays on the verify page only while its URL does. The specification, the vectors, the comparators, the rulings of record and the register are linked from this site the day their public URLs are named and answer without credentials; a repository is cited as evidence only once it is publicly reachable. The project's source is publicly reachable, read-only, with its log and every commit by hash: the tree of record at git.igneum.network, mirrored on GitHub. The list, with what each lets you check and what it does not, is on Verify it yourself.